ISO 50 and Dynamic Range

[mistaspeedy]

It seems that the ISO 50 setting on Canon cameras seems to give mixed results depending on the camera. Hopefully others can get similar positive results with their cameras as I have gained.

With some cameras there is INCREASED dynamic range compared to ISO 100.
Some cameras have exactly the same dynamic range as ISO 100.
Some cameras have LESS dynamic range than ISO 100.

You can take a look at this chart to see where your camera stands:
http://www.photonstophotos.net/Charts/PDR.htm

Before we even start this discussion, we need to differentiate two very different things:
1) RAW dynamic range of the image sensor when taking an identical exposure at ISO 50 and ISO 100 (NOT making the ISO 50 exposure twice as long/bright)
2) The camera metering system and how it chooses to expose images when you set ISO 50 or ISO 100.

I'm talking about situation #1.

It seems that my ancient 1D mark II is one of the fewer cameras to have a very slight boost in dynamic range at ISO 50.

VERIFYING WITH MY OWN TESTS:

https://giphy.com/gifs/26zzic5iyRLhTRTy0/fullscreen

Here are the same 100% crops as above, but in lossless PNG:
https://ibb.co/nMwxqQ ISO 50
https://ibb.co/ihcgH5 ISO 100

Here you can see the increased dynamic range of the ISO 50 image when both images were exposed identically.
That's an extra 0.18 stops according to photonstophotos.net

How I did the test:
1) Manually set the exposure of both images (ISO 50 and ISO 100) at F5.6 and 1/25 - purposely overexposing the whole image by about two stops (I want clipped highlights for the test, I don't shoot like this normally!)
*NOTE: The ISO 50 exposure is 0.7 stops darker than the ISO 100 image (not 1 stop!).
2) Reduce exposure of both images in adobe camera raw... ISO 50 image by 2 stops, and ISO 100 image by 2.7 stops to get equal brightness.
3) highlights slider set to -100 for both images.

Please note that both images have all the possible highlight detail recovered.... going to even more exposure compensation does not get more detail.

That's it... and you can see what the extra 0.18 stops gets you... a tiny bit more recoverable detail... hehe.

REAL WORLD APPLICATION:
The workflow for getting this extra dynamic range is quite simple, but goes against what the camera manufacturer has done, since Canon's aim with ISO 50 is less noise by overexposing by one stop.

1) Set -1.0 exposure compensation on camera (camera will meter as if shooting ISO 100).

2) When importing your images... just set exposure compensation to +0.7... and your exposure will match the brightness of a normal ISO 100 exposure. This one-click step can be set for hundreds of images at the same time... so there is no extra time spent.

WHAT ABOUT THE SHADOWS?

I have done various tests and can find no noise/color penalty/advantage of any kind by shooting ISO 50 vs ISO 100.


TOO LONG, DIDNT READ... & CONCLUSION:

Setting ISO 50 on my camera with the same exposure as ISO 100 gets me the following:

HIGHLIGHTS: 0.18 stops of extra highlight detail (increased dynamic range).
SHADOWS: No benefits and no disadvantages.... they are identical.


.... and if you're really bored... check the original raw files for yourself
http://www119.zippyshare.com/v/GqHFSvGQ/file.html
http://www119.zippyshare.com/v/fvJKEill/file.html

 

[neuroanatomist]

Expanded ISO settings are digital gain applied in camera. Thus, ISO 50 is a digital pull performed in-camera. However, even though the lowest native ISO setting for most cameras is ISO 100, the 'true' base ISO for most cameras is not ISO 100. Rather, it's usually somewhat lower and ISO 100 is actually a slight push. I suspect these slight differences in DR at ISO 50 arise from differences in the true base ISO (e.g. a pull from ISO 84 vs. a pull from ISO 76).

Incidentally, in some cameras the 'tweener' ISOs (e.g., 160, 250, etc.) are a digital pull or push, because analog gain is applied only in full stops, and that's why the DR curves are 'jagged' at the lower end.

 

[mistaspeedy]

All of what you said sounds right according to what I have read.
From what I can tell, those with the 1Ds Mark II seem to benefit the most... 0.24 stops, from using ISO 50.

The most popular discussion at the moment seems to be dynamic range for the 6D vs 6D mark II... with the original 6D seeing the tiniest measurable improvement from ISO 50, and the 6D mark II getting worse at ISO 50.

Long story short... 6D mark II dynamic range is a bit worse than the original 6D (until ISO ~2500), and there doesn't seem to be any significant high ISO dynamic range gains to offset this.

 

[timmy_650]

http://www.photonstophotos.net/Charts/PDR_Shadow.htm

Can someone explain to me on this chart why it looks like the 5Dm4 is a lot worse then the 5Dm3? Am i just reading it wrong?

 

[raptor3x]

http://www.photonstophotos.net/Charts/PDR_Shadow.htm

Can someone explain to me on this chart why it looks like the 5Dm4 is a lot worse then the 5Dm3? Am i just reading it wrong?

What you're seeing here is improvement in the input-referred noise relative to base ISO. They will always start at one since it's normalized but a lower value as you move up the curve either means that there was less noise at base ISO to improve upon, or you're doing a really terrible job of amplifying the signal at high ISO. Pretty much all manufacturers do well at high ISO amplification now so the chart is primarily a measure of how much noise there is at base ISO.

https://theory.uchicago.edu/~ejm/pix/20d/tests/noise/noise-p3a.html

 

[Valvebounce]

Hi mistaspeedy.
Thanks for sharing this, http://www.photonstophotos.net/Charts/PDR.htm it made for interesting reading. Unfortunately I have sold the only camera (5D) I had which has ISO 50, the 1DsIII is missing from the list.
Can someone answer a couple of questions for me please,
What do the numbers in this quote (from the 7DII, but each camera has its own number) mean? 'Low Light ISO Setting 1901.'
For this number, is bigger better or worse?

Cheers, Graham.

 

[mistaspeedy]

I assume that the bigger the number, the better.... it seems to make sense give how the cameras rank (smartphones worst, newer full frame and medium format best)

 

[Kit Lens Jockey]

This is really great info. I had always thought that ISO 50 was inferior to ISO 100 as far as dynamic range goes. Being that it's not in many new cameras, I have to question why it has to be set manually as opposed to being a value that auto ISO can use if it decides to.

 

[aceflibble]

Don't forget that subject matter and exposure time will also affect the dynamic range and noise you can perceive. Tests like this need to be done exceptionally carefully in order to prove anything. As a general rule I'd say people are best off checking DXO for their ISO sensitivity and SNR% measurements to see that ISO 50 is actually doing for any given camera, as DXO are as lab-strict as you can get and they have the largest catalogue of bodies to test.



Though for the record, there is no Canon camera which shoots ISO 50 natively. It is always the ISO 100 setting pulled by a stop after capture. What neuroanatomist said is mostly true other than that ISO 100 typically is not pushed, but simply left as-is at ISO 94, 80, or whatever else they're actually using for ISO 100. (ISO standards allow for some slight fudging of the exact figures, so if a camera shoots at ISO 80 the manufacturer can legally label it as ISO 100 without alteration. Canon went through a period of abusing this very badly, rating their ISOs as much as half a stop away from what they actually were. Fuji still do this. It's fine if you only use in-camera metering, but if you mix system or use a light meter, it's a real pain.)
Which means whether you'll get any benefit from it depends more on the processor (that's the processor in the camera, not the software on your computer) and how that specific model of camera has been configured to pull the exposure, rather than actual sensor dynamic range. This complicates testing and comparisons further as while everyone is quick to dive into every sensor in every camera ever sold, there are very limited resources detailing processor efficiency; people still can't decide if the old DIGIC II processors are actually altering the colour on purpose or produce the colour they do due to a fault.

Also your test up there is a bit iffy 'cause overexposing by two stops to test a single stop pull means you're blowing everything more than the feature is intended to compensate for. It's not unreasonable in real-world terms, of course—everyone is free to over- or under-expose as much as they like—just that for the sake of testing, you've taken things to more of an extreme which makes this result even less relevant for other systems than it already would be, given the huge change in processors since the 1D2. Not to mention using an outdoors scene isn't exactly a bullet-proof test subject.

 

[SecureGSM]

in afew words: the bigger the number the better camera performs at high ISO. this number corresponds with the highest "usable" (noise levels) ISO level for the camera. i.e. 5D III Low Light ISO is: 3652 and 5D IV Low Light number is: 5011. Than means that 5D III is as "noisy" at ISO 3652 as 5D IV at ISO 5011. or in another words: 5D IV has 1/3 stop high ISO advantage over the 5D III. I hope it does make sense.

What do the numbers in this quote (from the 7DII, but each camera has its own number) mean? 'Low Light ISO Setting 1901.'
For this number, is bigger better or worse?

Cheers, Graham.

 

[neuroanatomist]

Though for the record, there is no Canon camera which shoots ISO 50 natively. It is always the ISO 100 setting pulled by a stop after capture. What neuroanatomist said is mostly true other than that ISO 100 typically is not pushed, but simply left as-is at ISO 94, 80, or whatever else they're actually using for ISO 100. (ISO standards allow for some slight fudging of the exact figures, so if a camera shoots at ISO 80 the manufacturer can legally label it as ISO 100 without alteration.

Are you certain of that? For example, the 5DIII has a base ISO of 80, according to DxO. If you look at Bill Claff's read noise data for the 5DIII, you see that at lower ISOs the plot is 'jagged' in appearance, with read noise being lowest at ISO 160 (which has less noise that ISO 100), and lower at successive full-stop multiples of the base ISO (320, 640, etc.) than the adjacent ISO 100-200-400-etc., until the noise level rises to the point that the differences are lost. That seems consistent with the idea that multiples of ISO 100 are a 1/3-stop digital push and multiples of ISO 125 are a 2/3-stop digital push from the full-stop analog amplification from the base ISO 80.

 

[SecureGSM]

what Neuro just said is valid for nearly every Canon DSLR. I took the liberty of building Read Noise in DN vs ISO curve for 4 Canon bodies. image is attached. please note the copyright symbol in the right bottom corner of the chart.

 

[neuroanatomist]

what Neuro just said is valid for nearly every Canon DSLR. I took the liberty of building Read Noise in DN vs ISO curve for 4 Canon bodies. image is attached. please note the copyright symbol in the right bottom corner of the chart.

If you look back at the old 5D, which has a DxO base ISO of 92 (pretty close to 100), you see that multiples of ISO 100 have the lowest noise in Claff's data, with multiples of 160 having the highest noise (i.e. it appears that 125 multiples are the 1/3-stop push, and 160 multiples are the 2/3-stop push).

One exception to the 'valid for nearly every Canon DSLR' is the 1D X, which reportedly has separate analog amplifiers for each 1/3-stop progression, and thus a smooth curve vs. the jagged data resulting from digital pushing.

 

[GammyKnee]

One exception to the 'valid for nearly every Canon DSLR' is the 1D X, which reportedly has separate analog amplifiers for each 1/3-stop progression, and thus a smooth curve vs. the jagged data resulting from digital pushing.

Looks like the 5DS/R belongs on the exception list too.

 

[SecureGSM]

oh, I didn't know that. Thank you, Sir. No wonder 1D series bodies are so expensive.

what Neuro just said is valid for nearly every Canon DSLR. I took the liberty of building Read Noise in DN vs ISO curve for 4 Canon bodies. image is attached. please note the copyright symbol in the right bottom corner of the chart.

If you look back at the old 5D, which has a DxO base ISO of 92 (pretty close to 100), you see that multiples of ISO 100 have the lowest noise in Claff's data, with multiples of 160 having the highest noise (i.e. it appears that 125 multiples are the 1/3-stop push, and 160 multiples are the 2/3-stop push).

One exception to the 'valid for nearly every Canon DSLR' is the 1D X, which reportedly has separate analog amplifiers for each 1/3-stop progression, and thus a smooth curve vs. the jagged data resulting from digital pushing.

 

[Valvebounce]

Hi SecureGSM.
Thanks, that is a really clear and concise explanation to my rather ambiguous question!

Cheers, Graham.

in afew words: the bigger the number the better camera performs at high ISO. this number corresponds with the highest "usable" (noise levels) ISO level for the camera. i.e. 5D III Low Light ISO is: 3652 and 5D IV Low Light number is: 5011. Than means that 5D III is as "noisy" at ISO 3652 as 5D IV at ISO 5011. or in another words: 5D IV has 1/3 stop high ISO advantage over the 5D III. I hope it does make sense.

What do the numbers in this quote (from the 7DII, but each camera has its own number) mean? 'Low Light ISO Setting 1901.'
For this number, is bigger better or worse?

Cheers, Graham.